[mlpack] Plans for mlpack releases and features

[Ryan Curtin]

On Tue, Jan 01, 2019 at 02:20:21PM -0500, Tharindu Mathew wrote:
> Hi,
> 
> Happy new year to the mlpack community!
> 
> Overall, I'm trying to understand whether mlpack is actively developed (I
> understand it was an academic project taken up 10 years ago, and much has
> changed with everything (at least in the news) mostly being about deep
> learning now), or whether there's an active interest in adding more
> features and getting more releases out. I was wondering how one could
> contribute, ie whether there's a priority based feature/bug tracker.

Hey Tharindu,

Happy new year!

mlpack is definitely still actively developed---there is active PR
reviewing, merging, and responding to issues basically every day from
lots of different people.  The project has come a long way from its
origins at Georgia Tech as an academic project, like you pointed out.

There is not currently a formal release schedule but I would like to
have one (time has not permitted that though).  One of the big problems
right now is the difficulty of a release; it is a tedious process that
takes me something like 4+ hours of work, so, I have been working
towards an automated release process.  But time is short, so that isn't
done yet.

A lot of recent work has gone into splitting out the optimization
framework into a different project called ensmallen:

  https://github.com/mlpack/ensmallen

I can say that definitely there is active interest in adding more
features.  Here are some future directions that I intend to go, and that
I have heard others intend to go:

 * Automatically generated Markdown for each of the bindings.  I have
   most of this working now, but I need to get the Markdown->HTML parts
   working right.  See
   https://github.com/mlpack/mlpack/issues/1511.

 * GPU support via Bandicoot: https://gitlab.com/conradsnicta/bandicoot
   Personally this is high on my priority list and I am hoping in the
   next handful of weeks to start devoting more time to that.

 * A binding to the command-line or Python (or other languages) for
   neural networks: https://github.com/mlpack/mlpack/issues/1254

 * Callbacks for ensmallen optimizers:
   https://github.com/mlpack/mlpack/issues/1481

 * Bindings to Go (there is a PR open), Julia (I am nearly done with
   this), and R (I started but did not get far).

 * There are lots of mostly-finished PRs open that just need a little
   more to get them across the finish line.

 * Summer of Code 2019: we'll apply, although the organization
   application period isn't open quite yet.  A lot of work goes into
   GSoC every year.

That's not a comprehensive list for sure.  Others may have other things
that they are working on too. :)

I'm hoping to release mlpack 3.1.0 with better documentation, but I
think maybe in the next week or two I'll release mlpack 3.0.5 with some
backported bugfixes.  There's not really a formal process of any sort
for this, and if anyone else wanted to take over the process of
releasing the library and managing that I of course would have no issue
with it.

There's not really a prioritized bugtracker, although some Github issues
do have labels for priority.  Honestly I spend a significant portion of
my time just trying to stay above water with the different requests I
receive and the PRs that need to be reviewed. :)

I hope this helps!  I would say right now the mlpack community is the
most active it has ever been.

Thanks,

Ryan

-- 
Ryan Curtin    | "And do not attempt to grow a brain!"
ryan at ratml.org |   - Sgt. Howard Payne